This is a good paper! Though as a tool the contrast pattern used in the work is kind of too raw for such a complex problem, the idea of extracting STM patterns step by step is very interesting.